zai-org/CogVideo ? reverse-engineered prompt

Build me a video generation app based on CogVideo and CogVideoX that can take a text prompt, and for the newer model also an input image, then generate a short video and let me preview or save it.

I want a simple command line demo that is easy to run after installing the requirements, plus a clear path for both inference styles mentioned in the repo, the SAT version and the Diffusers version. Please make it work with the supported Python versions and include sensible defaults so someone can try it without knowing all the settings. If it helps, add a way to improve prompts before generation, since the model seems to work better with longer, clearer descriptions.

If the project supports it, also include the basic fine tuning flow so I can adapt the model to my own data later. Keep the setup straightforward, make the examples easy to understand, and update any code or instructions needed so the main demos run cleanly.